Qwen Image Edit Agent

Comprehensive AI image manipulation agent supporting Qwen models with advanced features for fashion e-commerce, character creation, and image enhancement.

Features

1. Qwen Image Editing (qwen_image_agent.py)

• Character consistency editing
• Outfit changes and body transformation
• Pose modification
• Background editing
• Single/multiple character support
• Async/await architecture for scalability
• Batch processing capabilities

2. Advanced Text Encoding (qwen_text_encode_advanced.py)

• Vision-Language (VL) processing with Qwen2.5-VL
• Support for multiple reference images (up to 3)
• Configurable megapixels for VL model (0.0-4.0 MP)
• Reference latent encoding via VAE
• Llama template integration for improved results
• Dynamic image scaling for optimal performance

3. Virtual Try-On & Body Shaper (virtual_tryon_body_shaper.py)

Virtual Try-On Features:

• Clothing try-on (tops, bottoms, dresses, shoes, jackets, full outfits)
• Multiple body shape support (Petite, Slim, Athletic, Curvy, Plus-Size, Tall)
• 8 camera angle presets:
  • Straight On / Front View
  • Side View / Profile
  • 3/4 View / Three Quarter Angle
  • Back View / From Behind
  • Overhead Shot / From Above
  • Low Angle / Hero Shot
  • Close Up / Macro
  • Full Body / Wide Angle
• Fit adjustment (-1.0 to 1.0: loose to tight)
• Length adjustment (-1.0 to 1.0: short to long)
• Custom prompt support
• CAD-Von & SDXL integration

Body Shaper Features:

• Body shape transformation (7 preset shapes + custom)
• Adjustable transformation strength (0.0-1.0)
• Face preservation option
• Natural skin tone and confident expression
• Professional photo quality enhancement

4. SDXL, LoRA & Character Creator (In Development)

SDXL Integration:

• Text-to-image generation with SDXL
• LoRA (Low-Rank Adaptation) model support
• LoRA weight control (0.0-2.0)
• Trigger word integration
• Multi-model inference pipeline

Character Creation:

• Consistent character generation
• Multi-pose variations
• Emotional expression control
• Style preservation across generations
• InstantID integration for face consistency

Quality Boosters:

• SUPIR upscaling (4x/8x resolution)
• Face detail enhancement
• Real-ESRGAN integration
• Latent upscaling options
• Dynamic quality adjustment (0.0-1.0)

Installation

# Clone the repository
git clone https://github.com/tabrezahmed51/qwen-image-edit-agent.git
cd qwen-image-edit-agent

# Install dependencies
pip install -r requirements.txt

Required Dependencies

httpx==0.24.1
aiofiles==23.2.1
pydantic==2.5.0
python-dotenv==1.0.0
pillow==10.1.0
numpy==1.26.2
aiohttp==3.9.1
tqdm==4.67.0
loguru==0.7.2

Usage Examples

Basic Qwen Image Editing

import asyncio
from qwen_image_agent import QwenImageAgent, ImageEditRequest, EditType

async def main():
    agent = QwenImageAgent(comfyui_endpoint="http://localhost:8188")
    
    request = ImageEditRequest(
        image_url="./input.png",
        edit_type=EditType.OUTFIT_CHANGE,
        prompt="Change to blue summer dress",
        steps=20,
        guidance_scale=7.5
    )
    
    result = await agent.process_edit_request(request)
    print(f"Result: {result}")

asyncio.run(main())

Virtual Try-On

from virtual_tryon_body_shaper import VirtualTryOnEngine, TryOnRequest, BodyShape, ClothingType, CameraAngle

engine = VirtualTryOnEngine()

request = TryOnRequest(
    model_image_url="./model.png",
    clothing_image_url="./dress.png",
    body_shape=BodyShape.CURVY,
    clothing_type=ClothingType.DRESS,
    camera_angle=CameraAngle.STRAIGHT_ON,
    adjust_fit=0.2,
    custom_prompt="elegant evening wear"
)

result = await engine.process_try_on(request)

Body Shape Transformation

from virtual_tryon_body_shaper import BodyShapeRequest

request = BodyShapeRequest(
    image_url="./person.png",
    target_body_shape=BodyShape.ATHLETIC,
    adjustment_strength=0.8,
    preserve_face=True
)

result = await engine.process_body_shape_transformation(request)

API Endpoints (Future)

• POST /api/edit - Process image editing request
• POST /api/tryon - Virtual try-on request
• POST /api/reshape - Body shape transformation
• POST /api/character - Character creation
• POST /api/batch - Batch processing

Model Requirements

ComfyUI Setup:

• ComfyUI (latest version)
• CUDA-capable GPU (8GB+ VRAM recommended)

Model Files:

• Qwen2.5-VL (vision-language model)
• SDXL base model
• LoRA models (optional)
• SUPIR model (for upscaling)
• Real-ESRGAN model (for enhancement)

Performance Notes

• Qwen Image Editing: ~30-60s per image (20 steps)
• Virtual Try-On: ~45-90s per image
• SUPIR Upscaling: ~60-120s per image
• Batch Processing: Scales linearly with number of concurrent requests

Architecture

qwen-image-edit-agent/
├── qwen_image_agent.py                    # Main async agent
├── qwen_text_encode_advanced.py            # Advanced text encoding with VL
├── virtual_tryon_body_shaper.py            # Try-on and body shaping
├── requirements.txt                        # Dependencies
└── README.md                               # This file

Supported Edit Types

1. CHARACTER_CONSISTENCY - Maintain character features across edits
2. OUTFIT_CHANGE - Change clothing while keeping character
3. POSE_CHANGE - Modify body pose and stance
4. BACKGROUND_CHANGE - Replace or modify background
5. MULTI_CHARACTER - Edit multiple people in one image
6. FULL_EDIT - Complete image regeneration with custom prompt

Supported Body Shapes

• Petite (small frame, shorter height)
• Slim (slender, lean physique)
• Athletic (muscular, toned)
• Curvy (hourglass figure)
• Plus Size (fuller figure)
• Tall (long legs, height emphasis)
• Custom (user-defined proportions)

Camera Angles

8 professional photography angles optimized for fashion e-commerce:

• Straight On / Front View
• Side View / Profile
• 3/4 View (Three Quarter Angle)
• Back View (From Behind)
• Overhead Shot (From Above)
• Low Angle (Hero Shot)
• Close Up (Macro)
• Full Body (Wide Angle)

Contributing

Contributions welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Submit a pull request

License

MIT License - See LICENSE file

Author

tabrezahmed51 - Full-stack developer specializing in AI/ML and ComfyUI workflows

Support

For issues, questions, or feature requests:

• GitHub Issues: Create an issue
• Email: Contact via GitHub

Roadmap

• SDXL-specific character creator module
• LoRA training pipeline
• SUPIR upscaling integration
• REST API server
• Web UI dashboard
• Model caching optimization
• GPU memory optimization
• Multi-GPU support
• Docker containerization
• Production deployment guide